I always feel bad saying I spent the week just testing and learning new ways to make the ai do what I want, but look! I think this shows I made some progress. First of all this quad is what Tucker asked in the comments to see - wearing a t-shirt and no bra. Now I don't normally do requests but here I was like, that's not that demanding, I could see my spin on that, however I couldn't get the ai to do that reliably enough combined with other stuff I wanted from it, to be happy with the result. And now I managed, with some new tools and workflows I'm experimenting with!
As for the other gifs, the armless ones are standard but I actually managed to do a hand-walking hemi, which was hugely difficult. Sadly there I do not have much control over other specifics yet, I'm just glad I got some passable results so far.
TLDR: Me complaining about consistency and control over multiple specific details in AI is difficult but claiming I'm improving on that front + hemi handwalk early tests.